LenaVolkova/ArmenianAddresses

Dataset Card for Armenian Address Extraction Dataset

<!-- Provide a quick summary of the dataset. -->

This dataset consists of 2,271 records containing raw Armenian utility maintenance/planned outage announcement texts paired with structured address components extracted from them. The raw texts primarily originate from public announcements (such as those by the Electric Networks of Armenia) notifying the public of scheduled maintenance, and the dataset breaks down these notices into granular, queryable geographic attributes.

Dataset Details

Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

• —Curated by: Elena Volkova
• —Language(s) (NLP): Armenian (hy)
• —License: Open Database License (ODbL) or Creative Commons Attribution 4.0 International (CC BY 4.0)

Dataset Sources

Parsed data from public websites ena.am and veolia.am

Uses

<!-- Address questions around how the dataset is intended to be used. -->

Direct Use

<!-- This section describes suitable use cases for the dataset. --> This dataset is intended for training, fine-tuning, and evaluating machine learning models on tasks such as:

1. 1.Named Entity Recognition (NER) and Slot Filling for Armenian address components.
2. 2.Address Parsing (converting unstructured text addresses into structured JSON/CSV formats).
3. 3.Information Extraction (IE) from official government or utility announcements in the Armenian language.

Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->

• —Parsing addresses in languages other than Armenian.
• —Geocoding tasks (converting text to exact latitude/longitude coordinates) without external lookup databases, as coordinate details are not included in this dataset.
• —Extracting postal codes/ZIP codes, as they are generally not present in this source material.

Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

The dataset contains a single split of 2,271 rows sorted by the key column. Each record consists of the following 9 fields:

• —key (Integer): Unique identifier / group key matching the announcement event.
• —text (String): Raw Armenian announcement text detailing the maintenance schedule and affected locations.
• —region (String): The target province/state/region (e.g., Երևան or Արարատի մարզ).
• —cityname (String): The name of the city, town, or village.
• —citytype (String): The settlement type class (e.g., քաղաք for city, գյուղ for village).
• —district (String): The administrative district or neighborhood (e.g., Կենտրոն, Շենգավիթ).
• —street (String): The street name (e.g., Չարենցի փողոց).
• —bldnum (String): The specific building number.
• —bldlist (String): A list, range, or collection of affected building numbers (e.g., 23-31, 41-55, 87-103).

If there are several addresses in the 'text'-field, then there will be several strings for this text in the dataset with one structured address in each string. In the jsonl-file this strings are united to the one with 'text' and list of all structured addresses corresponding to the text.

Dataset Creation

Curation Rationale

<!-- Motivation for the creation of this dataset. -->

Public utility announcements in Armenia are frequently published as large blocks of unstructured text. This makes it difficult to query or filter alerts by a specific street, neighborhood, or building. Structuring this data helps build models that can automatically parse and categorize locations affected by service interruptions.

Source Data

<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->

• —Source Data Type: Public utility maintenance and service outage announcements.
• —Source Language: Armenian.

Data Collection and Processing

<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->

Data was gathered from official public utility notification feeds (such as the Electric Networks of Armenia - Հայաստանի էլեկտրական ցանցեր). Unstructured announcement texts were processed and segmented. Text cleanup included lowercasing and normalizing common Armenian address keywords (e.g. converting Մարզ to մարզ, Փողոց to փողոց, etc.) to maintain consistency across records.

Who are the source data producers?

<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->

The source data is produced by public utility companies and service providers operating in the Republic of Armenia.

Annotations [optional]

<!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->

Annotation process

<!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->

The address segments were annotated and structured using a dedicated local database curation and web interface tool. Annotators verified the extracted street names, building numbers, and districts to align them with standard Armenian geographic naming conventions (using references like OpenStreetMap databases).

Who are the annotators?

<!-- This section describes the people or systems who created the annotations. -->

Elena Volkova (dataset developer/curator).

Personal and Sensitive Information

<!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->

This dataset does not contain personal names, phone numbers, email addresses, or private individual records. It contains only public physical address structures and building designations that were published in open, public maintenance announcements.

Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

• —Domain Bias: Because the source texts are utility announcements, the wording, syntax, and address formats are highly specific to maintenance reports. Models trained on this dataset might not perform as well on general postal addresses, package shipping labels, or casual chat queries.
• —Geographic Bias: The data is heavily concentrated around major urban administrative areas (such as Yerevan and its districts) and municipal centers. Rural areas may have less granular street-level representation due to how outage notices are clustered.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Dataset Card Contact

Elena Volkova